S4 | E3 | How AI Turns Supply-Chain Chaos into Clarity with Ilya Levtov – Co-Founder @ Craft

Welcome to the Think Data podcast brought to you in partnership with Mydataworks. If you want to stay up to date with the latest breakthroughs and trends in the world of data and artificial intelligence, and if you're curious about some of the strategies that companies and founders use to launch data and AI products, then you're in the right place. Our aim is to bring together a diverse lineup of fantastic guests from the founders, through to accomplished leaders and product owners at some of the most fascinating data and AI companies worldwide. They will each offer you their own unique insight into what it takes to launch and scale a great data business. Thanks for tuning in and I hope you enjoy the episode. Welcome to the Think Data podcast and today I'm really pleased to welcome Ilya Levtov to the show. Ilya is the co-founder and CEO of Craft. They're a business that have been around for just under 10 years now, but they are super, super interesting because they're an AI-powered supply chain resilience platform. And it's a unique proposition because it helps companies make real smart decisions around their supply chain here and stay ahead of disruptions. So really, simply put, it helps businesses stay resilient and keep moving. So, yeah, really good to have you on, Ilya. I know we kind of also share in the same. country of birth. You're obviously in the States at the moment. So I'm really keen to kind of go back to that backstory of how did this all start? You obviously went Juilliard, Goldman, venture capital. When did you stop and decide, I'm going to become an entrepreneur? I'm going to build something?

Yeah, that's really interesting. For people, obviously, who have not come across you or Kraft before, in simple terms, what is Kraft? And what is the problem statement here?

Yeah, Kraft is a supplier intelligence platform. And I'll answer it also, I'll give a bit of background because it wasn't that to begin with. So the quick story is that when I first started, it was a situation of sort of a founder, you know, scratching your own right and what was it that I'd felt frustrated by for that entire sort of career span I described. It was almost every single day I needed to look up information about a company for one reason or another, whether I was investing in it at Venrock or selling to it or partnering with a business or even applying for a job somewhere. I found myself, you know, almost every day for one reason or another, looking up information on companies and thinking this isn't very good. You know, the public platforms that are out there, they weren't complete. They focused on only one cohort of companies. or only one type of data. And I said, can we build a completely comprehensive, call it source of truth on companies. And we started building crawlers and scrapers that would pull together data from companies on websites and then social media and putting together really the deepest, most complete profile of a company that we possibly could. That was sort of step one and went for a few years, including us getting very high page rank and trust from Google, who after a while was showing us in about a hundred million. organic search results per month, bringing over to about two and a quarter million people onto our website with absolutely no marketing spend. And these would be people that just sort of searched a little tidbit about where's that company headquartered or, you know, how many marketing jobs do they have open right now? And we were showing up high in the search results. And that led to that audience. Now, from there, we started to say, okay, wow, we need to monetize, right? We're a business. We've got to try and make some money. who who's the audience that looks most interesting and quite surprisingly it was supply chain people that were coming that was showing up there and it turned out we discovered with their help that that we'd built quite a powerful and useful resource for large enterprises and so the first was a large aerospace and defense manufacturer who got in touch and said yeah we sit here as a team of six people trying to keep track of what's going on with 20 000 companies at once which is basically our tier one direct supply chain. No, and it's a nightmare. And recently we've kept stumbling onto craft and we love your data and your supplier profiles. Do you have an API? And that was the light bulb. That's where it started. So from there, we started really focusing. That was about five years ago. And we haven't looked back. We've really gone into collecting and structuring and being the most comprehensive resource of information about companies pointed at large enterprise supply chains. Yeah. Both commercial enterprise and government. We're working with about 26 federal government agencies, including out of the Pentagon. And what we're doing for all of them is structuring this data and turning it into intelligence about their supply chain. Who are their suppliers? Where are they? How are they doing? What's happening with them? Who owns them? Who works with them? And, you know, everything you would want to know. And so part of the use case is risk management, you know, in an era of... this supply chain disruption and, you know, very, very major risks around cyber and geopolitics in particular. But not only that, the second sort of use case for the supplier intelligence is on the positive side, helping companies optimize their supply chain and have fingertip intelligence to go and negotiate a better, stronger partnership with that supplier.

Interesting.

And so that's also what we help companies do.

Yeah, it's interesting you touched on the government piece, because I'm guessing from a data, from a source of truth standpoint, it has to be accurate, right? Because obviously they're going into, you talk about risk, but how do you and Kraft de-risk yourself then? Because obviously you're getting this data, which obviously you're, you talk about the last five years, you've really gone narrow on this. And rightly so, but how do you make sure that the data you're getting is the right data?

Yeah, that's a great question. Because throughout our industry, data accuracy is...

Yeah, it's everything, right?

Yeah. And I'll be very, very candid and say that 100.0% data accuracy is, you know, it's this far from impossible, basically. The fact is, you know, things happen. There's almost no context in which you can get to that degree of precision, which means that it is a bit about, I think, as you ask... the risk management within the data. So what do you do to make it as close to 100% as possible? And further, when there is a gap that you can spot it rather than it's going there hidden and causing an incorrect answer. And so that's where there are a few techniques that we do. So first of all, we've been big proponents of human in the loop. And even in the age of AI, that still exists. And so while our crawlers and scrapers have now turned into agents out there, collecting data we've still got a very robust human in the loop operation and plan to continue to have that to do various things that you know at least up until today still only a human can do with very very high reliability in terms of checking data spot checking and and programmatic checking at high scale and just ensuring you know supervising agents basically now it's evolving rapidly but but we still believe that human in the loop is an important part of it now another very major part of the platform at this point is our partnerships with specialty data providers in major areas. So financial data, for ESG data, for cybersecurity, for geopolitical risk, for all of these different dimensions of risk that we monitor and help mitigate for our customers, we've got some of the most high quality, best in class, best in the market data partners that just focus on that And in every data domain, we don't just have one, we have multiple. Now, what that enables us to do, this multi-sourcing becomes a very interesting mitigation for inaccuracy as well. Because if you've got three or four sources that are all saying the same thing from different angles, well, there your confidence went up. But in a case where, you know, they're diverging and they're saying, you know, a different point, well, that in itself is an interesting flag. And then you can send your human in to go and say, all right, well, one of these two is wrong. Let's figure out which one it is. And so a large part of the solution really is a kind of triage to say, no, here's all the stuff that's with very high confidence, clean and clear, and you don't need to focus on it. But here's your like, you know, seven suppliers right now in your portfolio where we either know something's going on or we think something might be going on that warrants a closer look. And when you're dealing with tens or hundreds of thousands of suppliers, that triage and that ability to focus people.

attention in the right place is really really valuable yeah it's one of the first people to talk about the human in the loop piece which i think actually for the listeners here is everyone's going full hog on ai and it would ignore the displacement uh discussion but it's more around people's jobs become either easier or they're enabling them to do more because of ai but i think one thing you touched on which i like and people listening don't underestimate the value that a human has still in this process? And you touched on something, you said... for now and i think obviously agentic ai is just going to get more intelligent and more intelligent but having someone there as a human that can go hang on a minute this doesn't seem right i think super important it also gives you that differentiator against maybe other organizations saying oh we're pure ai how does the ai agents work across specifically because obviously they're being deployed to go and obviously data scrape when that comes in are you then you're obviously putting those different layers you're obviously pulling two or three sources are they Are they then almost like red flagging certain data points? They're going, you need to check this.

Yes, that's exactly what happens. So the agentic AI is, we've got it deployed in two main areas. So just to summarize exactly as you said, first of all, it's collecting whatever publicly available data out there. And then that's being collected along with specialty non-public data from the types of data partners. that I mentioned that are very specialized in those individual domains of data, such as cybersecurity or foreign ownership control and influence data and things like that. Now, when we've got the data in our data lake and... It's structured, some of it's unstructured. Then we've got various agents and the most important and prominent one is essentially a risk analyst agent. And what that agent is doing is consuming all of the data on a target supplier or a target portfolio of suppliers with one of our customers. And then absorbing, aggregating all of the data of every type and basically outputting an assessment. a report and that will go. So our AI risk reports that we launched earlier this year, very rapidly absorb all that data and come out with conclusions. And yes, it's basically flagging high, medium, low risk, which is configurable by our customers because high, medium, low risk means different things to different people. Well, exactly right. They've got quite stringent definitions and we've been working closely with them and 26 different agencies for over five years. really fine tuning and making our risk models answer very closely to their specific parameters. And that is another thing that differentiates us from competitors, the work that we've done there. And ultimately, so that risk analyst agent is then producing the report that then goes to the human analyst and can be amended, further prompt, configured, and sort of brought to perfection. Now to... Give an example of the time savings. We've gone from, you know, a couple of years ago, a human analyst was spending on average eight hours to do the deep dive research on one key supplier into the US industrial base. We're now outputting that report in about 30 to 40 seconds. And then, and we do say that there is another further, you know, half an hour or so where the consumer of that report, the analyst, the military analyst, is then spending time with it because they're not just believing that what the agent, the AI has said. Okay,

so that third layer.

Yeah, completely. And so for now, we basically say we've shrunk an eight-hour process down to one hour. You know, rough numbers. But that's, you know, it's an example, sort of like 85% ROI. And we've enabled, you know, a very high percentage of teams that were doing exclusively this work in a very manual way. We've empowered them, we've accelerated them, we've increased the consistency and the quality of the work using AI. But then we've also freed up massive amounts of time for them to go and do other things.

Well, that's it, isn't it? It's the freeing up of time. A lot of people do to wet displacement in certain sectors there is. But I had someone on a while back who runs contact center AI. And actually all he said was actually the step up into contact center roles now is just you're bypassing one or two jobs on the ladder. and going to more impactful work and doing more work that actually you probably want to be doing. It's the job you want to get to is actually just happening quicker. So I find that interesting. What I want to touch on now is you, right at the beginning, you mentioned about Google ranking, SEO. You were getting all the 2 million hits a month. There's all this volume coming in. How has your go-to-market motion changed over the years, especially in light of AI and the amount of data that's out there?

how accessible certain sites make data how's that go-to-market motion happening yeah yeah absolutely it was there's a product aspect to that question and then the and then the the go-to-market so first just to touch on the product you're right there's so much data out there that was one of the big tailwinds for the whole business idea from the beginning that we're sort of well into the digital age and lots of data that used to just not exist digital anywhere you know being in a file cabinet or in someone's brain was now sort of generally available to to absorb and make sense of and that's true. And the global supply chain is just one of the biggest and, in our view, most fascinating data sets out there. You know, a vast collection of nodes and connections, basically, that makes up the global supply chain. And it's an absolutely vast trove of data. And really, AI is the only method through which you could ever hope to really kind of understand it. And now in this new era of AI. That's all possible in ways that never was before. So that's very exciting in terms of how data has changed. Data availability has changed what's possible. On the go-to-market side, what we found, and I think this goes to a slightly traditional answer, where we just found that it really was enterprise sales that we were getting into. because we serve some of the largest... companies in the world, you know, Fortune 100 and Fortune 500 size and the equivalent in the government sector. And so those processes very much are, you know, led by experienced, seasoned account executive that can get in close. with that customer, build a trusted relationship, understand how they are running these types of processes today, bring a solution engineer along with them, do some whiteboarding and powerfully demonstrate how with craft technology added into their process, things get better and more efficient in a large number of ways. So I think that's sort of fairly traditional and there's no magic to that. What we are seeing now, though, is with LLMs and more data out there, it's possible to just understand the prospect universe way better than what's possible.

You can get deeper, can't you? You can get far deeper now.

Yeah. And so what we're doing there is using that to inform, well, who should we actually go and talk to? Because who has got a process that's in that right step of maturity, not at zero, but not also super evolved, where we can go in and have highest impact and have the highest. likelihood of within a reasonable period of time successfully getting a deal getting them to sign on the dotted line and becoming a customer which is at the end of the day what it's all about so using data and ai to inform that is is absolutely part of improvements we've we've seen in the last couple of years in the sales and marketing yeah

again it goes back to your point else is ai is the enabler right it's not it's not ai to replace and actually there's too much talk about ai sdrs and ai outreach but the reality is no better than a person making a either a call or an outreach on email that's relevant tailored they understand the buyer persona they've got access to real-time data and they can go this could probably save you time and money it's people as a as a receiver of that call you're gonna be like oh okay i'm probably gonna listen to that as opposed to all

the nonsense we see in the world at the moment about it's just ai everything yeah completely i mean it's just so it's so hyped and there's an awful lot of slop And yeah, at every function, we just really still believe in quality and also the human connection. We've started using the term advanced intelligence, which is a little bit of a play on AI. We're about to push out a website upgrade next week, and that's prominent in the heading because it's a little bit of a nod to, well, why does it have to be artificial? You know, some of it can be, but. But, you know, there's still a lot of value in the human connection. And we love working closely with our customers as partners, you know, a real spirit of design partnership. Because this product problem is complex and different at different companies. I talked about configuring risk models differently. And that's only the beginning. At the end of the day, optimizing supplier evaluation and qualification and continuous monitoring for a large enterprise is a bit different in. different companies. It's just going to be, and we really embrace that and say, we're going to go in and make it, you know, really configured and closely tailored to what you need. And while we're doing that, we're going to be learning a lot and then cross-pollinating best practices across customers. So you heard this term, service as a software, which is coming up now, right? And I think it's a nod to that. It's a nod to the idea of forward deployed engineers, something that Palantir really pioneered. And it says, we're going to go on site. We're going to get really, really close to you, understand the ins and outs of your situation and give you something really closely tailored to what you need. We love doing that.

Yeah, I think it's super important. And I think you've also got that experience, the years, the knowledge that, as you say, is in here. And, you know, it's hard to unpick that and send that. via an AI or over an email. I think what I'd love to close things off with is what's next? Because you're obviously, in the last five years, you said it yourself, you've gone really deep into supply chain. You've got an amazing network of customers. You're very close to them. You're evolving very quickly. Do you see yourself, and maybe it's something you can publicly say or not, do you see yourself as just doubling down and becoming even well, better known in supply chain? Or do you actually see some new data categories happening and some opportunities elsewhere?

Yeah, I think for now, you know, for the next few years, all the foreseeable future, we've got absolutely more than our hands full in supply chain and even more focused category there is of procurement and really helping procurement leaders as part of the overall supply chain practice really optimize all of their decisions around suppliers from who are those suppliers and then how are they managing them on a continuous basis through time. So tons and tons and tons still to be done there. you know, a somewhat ultimate goal is to have the definitive map of the entire global supply chain. Right. That's just sort of like the intellectual challenge. And eventually, I don't think I wouldn't rule it out, broadening from supply chain intelligence to other functions in the enterprise and really deliver at the end of the day, very, very advanced enterprise intelligence in other functions besides supply chain. But I would have to say that for the next five years, my best guess is that we're more than occupied really optimizing supply chains and maybe longer than that, maybe forever. No shortage of fun.

Yeah, I think it plays, I'm a massive believer. It's being known for something and being great at something as opposed to, it's very easy I think in this market to see the next shiny object and shiny opportunity to just become a jack of all trades, master of none. But what you've done really well and hence why you're still going in such a competitive space and have doubled down and have these, you know, to get into these, what, 26 government. offices within the pentacle that's no mean feat and actually that's because you're becoming known for something as opposed to saying well you also do healthcare data you also do financial data and actually that's just not relevant to us that's exactly right it's focus isn't it the end of the day important uh asset a startup has its focus and uh we've been increasing the focus and narrowing it and we're we're going to keep doing it amazing ilia it's been loads of fun it's great to have you on and i feel we should uh almost get a second follow-up one booked in uh As soon as there's loads to unpick here. But I think for anyone listening here, there's a new website you said launching in just over a week.

Yeah, craft.com.

Perfect. Everyone should check that out. Obviously, I've not seen the new one yet, but I'm sure it'd be great. And then when we go live, I'm assuming people can take a look at the product. Is there some case studies? Is there some things they can play around with?

That's exactly right. And if you're in the supply chain space and looking to optimize your supplier network, click that, schedule a demo, and we'd love to talk with you.

That's it. Ilya, thanks so much this morning. It's been lots of fun.

Thank you very, very much. Again, cheers. Bye.